Longitudinal firm-level data: problems and solutions

Empirical measures of firm and employment dynamics based on administrative datasets are biased due to missing links in the longitudinal observation of firms. This paper presents a systematic overview of the problems and evaluates two prevailing solutions. We quantify the biases in a set of widely used empirical measures and show which estimates are most sensitive to missing linkages. The biases are found to be especially large in the size distribution of entrants and exits, in firm-level growth estimates for medium and large firms, and in job reallocation measures. We show that an employee-flow linkage method is more effective in reducing bias than a traditional link method often used by statistical agencies. A consistent approach is developed for imputing firm-level growth measures of linked firms. The analysis is carried out using a longitudinal dataset for Belgium and discussed from an international perspective.